How to train model for more than 2 speakers?

[Sonish-Maharjan-2014]

I trained the model (form NBSS) branch for 2 speakers separation using wsj0 dataset. It perfectly worked. But now I want to train the model for more than 2 speakers. What steps should I follow?

[quancs]

Hello, thank you for your insterests in our works. To train the model for a dataset where each utterance more than 2 speakers, you can change the number of output channels to 2N (N speakers and each has 2 numbers for the real and imginary parts of STFT coefficients) for each TF-bin.

[Sonish-Maharjan-2014]

Thank you for your response.. I tried adapting the code for four speakers. I generated room impulse responses (RIR) for the four speakers and made some adjustments in the code. Unfortunately, I ran into an error towards the end of the process.

Could you help me fix the problem?

[quancs]

You can debug your code to check the shape of echoics, echoic_i, and the value of needed_lens

[Sonish-Maharjan-2014]

Thanks, I will try to debug